Can memory deocder be used for vlm?

[MEKSAAA]

Hi, I'd like to ask if your memory decoder can be adapted to vision language model (e.g., Qwen-vl) ？I noticed that your experiments are based on language model.

[mxmcao]

Hi, I'm not very familiar with vlm in general. I do believe that this architecture can be easily adapted to vlm. However in order for this method to work, knn-search based method needs to have superior result first, which I can't confirm in the domain of vlm.